Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/MIRToolbox/mirnovelty.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function varargout = mirnovelty(orig,varargin)
wolffd@0 2 % n = mirnovelty(m) evaluates the novelty score from a similarity matrix.
wolffd@0 3 % [n,m] = mirnovelty(m) also return the similarity matrix.
wolffd@0 4 % Optional argument:
wolffd@0 5 % mirnovelty(...,'Distance',f) specifies the name of a dissimilarity
wolffd@0 6 % distance function, from those proposed in the Statistics Toolbox
wolffd@0 7 % (help pdist).
wolffd@0 8 % default value: f = 'cosine'
wolffd@0 9 % mirnovelty(...,'Similarity',f) specifies the name of a similarity
wolffd@0 10 % measure function. This function is applied to the result of the
wolffd@0 11 % distance function. cf. mirsimatrix
wolffd@0 12 % default value: f = 'exponential'
wolffd@0 13 % corresponding to f(x) = exp(-x)
wolffd@0 14 % mirnovelty(...,'KernelSize',s) or more simply mirnovelty(...,s)
wolffd@0 15 % specifies the length of the gaussian kernel, in samples.
wolffd@0 16 % default value: s = 64.
wolffd@0 17 % mirnovelty(...,'Normal',0) does not normalize the novelty curve
wolffd@0 18 % between the values 0 and 1.
wolffd@0 19 %
wolffd@0 20 % Foote, J. & Cooper, M. (2003). Media Segmentation using Self-Similarity
wolffd@0 21 % Decomposition,. In Proc. SPIE Storage and Retrieval for Multimedia
wolffd@0 22 % Databases, Vol. 5021, pp. 167-75.
wolffd@0 23
wolffd@0 24 dist.key = 'Distance';
wolffd@0 25 dist.type = 'String';
wolffd@0 26 dist.default = 'cosine';
wolffd@0 27 option.dist = dist;
wolffd@0 28
wolffd@0 29 sm.key = {'Measure','Similarity'};
wolffd@0 30 sm.type = 'String';
wolffd@0 31 sm.default = 'exponential';
wolffd@0 32 option.sm = sm;
wolffd@0 33
wolffd@0 34 K.key = {'KernelSize','Width'};
wolffd@0 35 K.type = 'Integer';
wolffd@0 36 K.default = 64;
wolffd@0 37 option.K = K;
wolffd@0 38
wolffd@0 39 transf.type = 'String';
wolffd@0 40 transf.default = 'TimeLag';
wolffd@0 41 transf.choice = {'Horizontal','TimeLag'};
wolffd@0 42 option.transf = transf;
wolffd@0 43
wolffd@0 44 normal.key = 'Normal';
wolffd@0 45 normal.type = 'Boolean';
wolffd@0 46 normal.default = 1;
wolffd@0 47 normal.when = 'After';
wolffd@0 48 option.normal = normal;
wolffd@0 49
wolffd@0 50 specif.option = option;
wolffd@0 51 specif.combineframes = @combineframes;
wolffd@0 52 specif.nochunk = 1;
wolffd@0 53 varargout = mirfunction(@mirnovelty,orig,varargin,nargout,specif,@init,@main);
wolffd@0 54
wolffd@0 55
wolffd@0 56 function [x type] = init(x,option)
wolffd@0 57 type = 'mirscalar';
wolffd@0 58 if not(isamir(x,'mirscalar') && strcmp(get(x,'Title'),'Novelty'))
wolffd@0 59 x = mirsimatrix(x,'Distance',option.dist,'Similarity',option.sm,...
wolffd@0 60 'Width',option.K,option.transf);
wolffd@0 61 end
wolffd@0 62 if isa(x,'mirdesign')
wolffd@0 63 x = set(x,'Overlap',ceil(option.K));
wolffd@0 64 end
wolffd@0 65
wolffd@0 66
wolffd@0 67 function y = main(orig,option,postoption)
wolffd@0 68 if iscell(orig)
wolffd@0 69 orig = orig{1};
wolffd@0 70 end
wolffd@0 71 if not(isa(orig,'mirscalar'))
wolffd@0 72 s = get(orig,'Data');
wolffd@0 73 dw = get(orig,'DiagWidth');
wolffd@0 74 for k = 1:length(s)
wolffd@0 75 if isnumeric(dw)
wolffd@0 76 dwk = dw;
wolffd@0 77 else
wolffd@0 78 dwk = dw{k};
wolffd@0 79 end
wolffd@0 80 if option.K
wolffd@0 81 cgs = min(option.K,dwk);
wolffd@0 82 else
wolffd@0 83 cgs = dwk;
wolffd@0 84 end
wolffd@0 85 cg = checkergauss(cgs,option.transf);
wolffd@0 86 disp('Computing convolution, please wait...')
wolffd@0 87 for z = 1:length(s{k})
wolffd@0 88 sz = s{k}{z};
wolffd@0 89 szm = max(max(sz));
wolffd@0 90 for i = find(isnan(sz))
wolffd@0 91 sz(i) = szm;
wolffd@0 92 end
wolffd@0 93 cv = convolve2(sz,cg,'same');
wolffd@0 94 nl = size(cv,1);
wolffd@0 95 nc = size(cv,2);
wolffd@0 96 if nl == 0
wolffd@0 97 warning('WARNING IN NOVELTY: No frame decomposition. The novelty score cannot be computed.');
wolffd@0 98 score{k}{z} = [];
wolffd@0 99 else
wolffd@0 100 sco = cv(floor(size(cv,1)/2),:);
wolffd@0 101 incr = find(diff(sco)>=0);
wolffd@0 102 if not(isempty(incr))
wolffd@0 103 decr = find(diff(sco)<=0);
wolffd@0 104 sco(1:incr(1)-1) = NaN(1,incr(1)-1);
wolffd@0 105 if not(isempty(decr))
wolffd@0 106 sco(decr(end)+1:end) = NaN(1,length(sco)-decr(end));
wolffd@0 107 end
wolffd@0 108 incr = find(diff(sco)>=0);
wolffd@0 109 sco2 = sco;
wolffd@0 110 if not(isempty(incr))
wolffd@0 111 sco2 = sco2(1:incr(end)+1);
wolffd@0 112 end
wolffd@0 113 decr = find(diff(sco)<=0);
wolffd@0 114 if not(isempty(decr)) && decr(1)>2
wolffd@0 115 sco2 = sco2(decr(1)-1:end);
wolffd@0 116 end
wolffd@0 117 mins = min(sco2);
wolffd@0 118 rang = find(sco>= mins);
wolffd@0 119 if not(isempty(rang))
wolffd@0 120 sco(1:rang(1)-1) = NaN(1,rang(1)-1);
wolffd@0 121 sco(rang(end)+1:end) = NaN(1,length(sco)-rang(end));
wolffd@0 122 end
wolffd@0 123 end
wolffd@0 124 score{k}{z} = sco;
wolffd@0 125 end
wolffd@0 126 end
wolffd@0 127 end
wolffd@0 128 else
wolffd@0 129 score = get(orig,'Data');
wolffd@0 130 end
wolffd@0 131 if not(isempty(postoption)) && postoption.normal
wolffd@0 132 for k = 1:length(score)
wolffd@0 133 for l = 1:length(score{k})
wolffd@0 134 sco = score{k}{l};
wolffd@0 135 sco = (sco-min(sco))/(max(sco)-min(sco));
wolffd@0 136 score{k}{l} = sco;
wolffd@0 137 end
wolffd@0 138 end
wolffd@0 139 end
wolffd@0 140 n = mirscalar(orig,'Data',score,'Title','Novelty');
wolffd@0 141 y = {n orig};
wolffd@0 142
wolffd@0 143
wolffd@0 144 function old = combineframes(old,new)
wolffd@0 145 if not(iscell(old))
wolffd@0 146 old = {old};
wolffd@0 147 end
wolffd@0 148 if not(iscell(new))
wolffd@0 149 new = {new};
wolffd@0 150 end
wolffd@0 151 for var = 1:length(new)
wolffd@0 152 ov = old{var};
wolffd@0 153 nv = new{var};
wolffd@0 154 ofp = get(ov,'FramePos');
wolffd@0 155 ofp = ofp{1}{1};
wolffd@0 156 nfp = get(nv,'FramePos');
wolffd@0 157 nfp = nfp{1}{1};
wolffd@0 158 od = get(ov,'Data');
wolffd@0 159 od = od{1}{1};
wolffd@0 160 onan = find(isnan(od));
wolffd@0 161 od(onan) = [];
wolffd@0 162 ofp(:,onan) = [];
wolffd@0 163 nd = get(nv,'Data');
wolffd@0 164 nd = nd{1}{1};
wolffd@0 165 nnan = find(isnan(nd));
wolffd@0 166 nd(nnan) = [];
wolffd@0 167 nfp(:,nnan) = [];
wolffd@0 168 [unused omatch nmatch] = intersect(ofp(1,:),nfp(1,:));
wolffd@0 169 if isempty(omatch)
wolffd@0 170 ov = set(ov,'FramePos',{{[ofp nfp]}},'Data',{{[od nd]}});
wolffd@0 171 else
wolffd@0 172 lm = length(omatch);
wolffd@0 173 ov = set(ov,'FramePos',{{[ofp(:,1:omatch(1)-1) nfp]}},...
wolffd@0 174 'Data',{{[od(1:omatch(1)-1),...
wolffd@0 175 (od(omatch).*(lm:-1:1) + nd(nmatch).*(1:lm))/(lm+1),...
wolffd@0 176 nd(nmatch(end)+1:end)]}});
wolffd@0 177 end
wolffd@0 178 old{var} = ov;
wolffd@0 179 end
wolffd@0 180
wolffd@0 181
wolffd@0 182 function y = checkergauss(N,transf)
wolffd@0 183 hN = ceil(N/2);
wolffd@0 184 if strcmpi(transf,'TimeLag')
wolffd@0 185 y = zeros(hN,N);
wolffd@0 186 for j = 1:N
wolffd@0 187 for i = 1:hN
wolffd@0 188 g = exp(-((i/hN)^2 + (((j-hN)/hN)^2))*4);
wolffd@0 189 if j>hN && j<hN+i
wolffd@0 190 y(hN-i+1,j) = -g;
wolffd@0 191 else
wolffd@0 192 y(hN-i+1,j) = g;
wolffd@0 193 end
wolffd@0 194 end
wolffd@0 195 end
wolffd@0 196 else
wolffd@0 197 y = zeros(N);
wolffd@0 198 for i = 1:N
wolffd@0 199 for j = 1:N
wolffd@0 200 g = exp(-(((i-hN)/hN)^2 + (((j-hN)/hN)^2))*4);
wolffd@0 201 if xor(j>i,j>N-i)
wolffd@0 202 y(i,j) = -g;
wolffd@0 203 else
wolffd@0 204 y(i,j) = g;
wolffd@0 205 end
wolffd@0 206 end
wolffd@0 207 end
wolffd@0 208 end